Possible design-induced artifacts associated with designs for discrete choice experiments

Abstract

Discrete choice experiments (DCEs) are widely used in many areas of applied social science research. The results of DCEs depend on the particular experimental design for the identification of the key parameters of interest and the statistical efficiency with which those parameters are estimated. Work on experimental designs for DCEs has almost always assumed that the particular design one uses does not influence the nature of the responses to the choice tasks other than via the precision with which parameters are estimated. We examine this assumption by testing whether particular experimental designs influence the probability that a separating hyperplane exists that perfectly predicts the observed choices at the individual level in four DCE data sets. Our empirical results suggest that the particular statistical design used can influence the nature of the choice responses obtained.